mescla.emma.archetypes.archetypal_analysis

mescla.emma.archetypes.archetypal_analysis(X, n_archetypes, *, n_restarts=10, seed=0, max_iter=200, tol=1e-06, standardize=True, standardizer=None)[source]

Fit n_archetypes hull points to water samples.

Parameters:

  • X (WaterChemistry, DataFrame or array) – Mixed samples. Candidate end-members do not belong here, for the same reason they are kept out of mescla.emma.model.EMMA.fit(): they would help define the hull they are supposed to be judged against.

  • n_archetypes (int) – Usually EMMA.n_endmembers, i.e. k + 1. Compare ranks with archetype_rss_curve() rather than assuming.

  • n_restarts (int, default 10) – Restarts from different initialisations; the best objective wins and the spread across restarts is reported. The objective is not convex, so this is not optional decoration.

  • seed (int, default 0) – Fixed so a fitted result is reproducible. It is recorded in the result.

  • standardize (bool, default True) – Z-score first, as EMMA does, so a large-concentration species does not define the hull on its own.

  • standardizer (Standardizer, optional) – A standardiser already fitted on these samples. Supply it to keep one standardisation across an analysis.

  • max_iter (int)

  • tol (float)

Returns:

ArchetypeResult

Return type:

ArchetypeResult

Notes

The fit is a candidate-generation step. See the module docstring for the four cautions that come with it; ArchetypeResult.summary() restates them.

Examples

>>> res = archetypal_analysis(samples, 3)
>>> print(res.summary())
>>> res.archetypes_frame()                              # candidate compositions
>>> ratios = mixing_ratios(res.as_endmembers(), samples) # ratios come from here